# Pausing AI Would Be a Mistake

Published: 2026-11-04

By [Ron Edgecomb](/about/index.md)

Tags: AI, AI policy

Today, Congresswoman Yassamin Ansari [posted on X](https://x.com/RepYassAnsari/status/2118054336077127797):

> Top of the Democratic agenda on day 1 needs to be a pause on advanced AI development, regulation of this out of control industry, and a ban on artificial superintelligence.

I think making a pause in AI development a day-one priority would be a serious mistake. I'm very much in favor of continuing to develop AI, and I want us to build the datacenters that make that possible. The more I see these systems contributing to work in science, engineering, mathematics, and security, the harder it is to accept that stopping further progress should be a political goal.

I had a similar reaction to Abdul El-Sayed's recent _60 Minutes_ interview. Asked about limiting work at the frontier of AI development, he called for a ["full on pause"](https://www.youtube.com/watch?v=YX44XDt0gXI&t=327s). He [raised concerns about rapid economic disruption, bioweapon misuse, and losing human control of AI](https://www.youtube.com/watch?v=YX44XDt0gXI&t=82s). He also proposed [state and local datacenter moratoriums](https://www.youtube.com/watch?v=YX44XDt0gXI&t=290s) until there are federal rules covering community agreements, utility costs, water, grid reliability, and labor.

Yes, it's reasonable to have concerns about employment, household bills, and corporate accountability. People should be able to have a say in what gets built in their communities. El-Sayed also raises the question of whether AI's benefits will actually reach the people who need them. That's a fair question to ask, but I disagree with making a halt in development the answer.

His [published platform](https://abdulforsenate.com/2026/10/ai-under-control/) goes a step further, calling for development to stop until safeguards ensure that "all risks to human safety are eliminated." It's unreasonable to expect every possible risk to be eliminated before a technology can advance. What we need are standards that let us evaluate particular risks, require protections, and hold people accountable. Evidence of dangerous behavior should justify restrictions on a system, and we should keep doing rigorous safety research as we improve the models themselves. If the condition for moving forward is a promise of absolute safety, it's hard to see how development would ever resume.

What frustrates me most is how narrow the public picture of AI can feel. If someone's main encounters with it are a chatbot giving an unhelpful answer, a bad search summary, or a feed full of generated images, I can understand why they might question the enthusiasm. Their personal experience gives them a very incomplete view of what these systems are already helping people accomplish.

On November 3, OpenAI [described using its models to design and build physical hardware](https://openai.com/index/accelerating-engineering-for-the-physical-world/). Two team members worked with Astra to design a device's enclosure, circuit board, firmware, and software integrations in under a week. A manufacturer produced a small batch of prototypes in under a month. The design still needed review and revision because the model hadn't consistently accounted for the manufacturer's production limits. I find it exciting that a small team could get from requirements to a manufactured prototype that quickly.

There's similar work happening in biology. Anthropic recently [reported work with six organizations developing new medicines](https://www.anthropic.com/research/early-progress-drug-development). One of those teams, Generate:Biomedicines, used Claude and other computational tools to [redesign part of a receptor used in CAR T-cell research](https://generatebiomedicines.com/blog-posts/less-is-more-refactoring-the-car-signaling-domain/). Their design shortened its signaling component by about 70%, with laboratory tests showing improved receptor expression and tumor-cell killing. It's early work, and benefits for patients still need to be established, but those lab results give researchers something promising to build on.

In mathematics, Anthropic [released a complete computer-checked proof of Fermat's Last Theorem](https://www.anthropic.com/research/formalizing-fermats-last-theorem), produced largely autonomously by Claude over 11 days. Claude converted an established human proof into Lean so a computer could check each logical step. Checking a complicated mathematical argument is difficult work in its own right, so making that process more practical could help mathematicians assess new results and build on them with greater confidence.

And even today, OpenAI [reported a research pipeline that reproduced 1,724 vulnerabilities in the Linux kernel](https://openai.com/index/a-thousand-fewer-paths-to-root/). It also proposed patches, with human researchers reviewing the findings and working with maintainers on reports and fixes. An enormous amount of computing infrastructure runs on Linux, so that work can help people who might never interact with an AI model themselves.

That security work also illustrates why the risks need serious attention. Systems capable of finding vulnerabilities can help attackers too, which makes better defensive tools, careful handling of those capabilities, and expert review especially important.

Of course, these companies want people to see their products in a positive light. I think we can keep that in mind while still recognizing what they and their users have accomplished. We have manufactured prototypes, laboratory experiments, computer-checked proofs, and security flaws reproduced in controlled environments to examine. We can ask what worked, what failed, and how much the systems actually contributed.

I want that evidence to carry more weight in the political conversation. A pause on new development wouldn't make every existing AI application disappear. It would restrict work on the next generation of systems, including capabilities that could make this kind of research more useful and accessible. I think the progress we could delay or lose deserves just as much attention as the risks a pause is supposed to address.

A U.S. pause also wouldn't mean a pause in China. El-Sayed [argues that much of China's frontier work imitates American models](https://www.youtube.com/watch?v=YX44XDt0gXI&t=340s), but Chinese labs are already [producing competitive models](https://hai.stanford.edu/ai-index/2026-ai-index-report) and [developing new architectures](https://deepseek.com/en/news/v4-preview/). Without an international agreement, I don't think we can count on that work slowing down just because we choose to stop.

This is also why I'm in favor of building more datacenters. Training and running capable models requires physical computing infrastructure, and the research I've been describing depends on access to it. I want us to build that capacity, along with the energy infrastructure needed to support it.

Building that capacity takes planning. The International Energy Agency [projects that global datacenter electricity use will roughly double between 2025 and 2030](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary), and describes real bottlenecks in electricity supply and grid connections. As demand grows, we need to invest in generation, transmission, storage, efficiency, and better coordination between developers and utilities. Companies should pay for the infrastructure their projects require, and decisions about where to build should take local water supplies and residents' quality of life seriously.

Ohio offers one example of how we can do that. In July 2025, [Ohio's utility regulator ordered AEP Ohio to create a datacenter-specific tariff](https://content.govdelivery.com/accounts/OHPUC/bulletins/3e8bb79) intended to protect other customers from the costs of underused infrastructure. The same order required AEP to lift its moratorium on new datacenter connections. That doesn't settle every question about where to build, but it shows how we can put protections in place and still allow development to proceed.

As I wrote in [my last note](/notes/hate-slop-not-ai/), I think we get further by being specific about what's causing harm. If a company is deceiving people, deploying a system irresponsibly, treating workers poorly, or passing its costs on to a community, those are specific problems we can address. Treating AI development itself as the problem sweeps a great deal of worthwhile work into the same category.

I understand why people are angry about some of what the industry produces and how powerful companies behave. I also want the conversation to make more room for the researchers, engineers, developers, and smaller teams using these tools to do valuable work. Their work is part of what a pause would put at stake.

I want more people to have access to capable AI, more scientists to be able to test ambitious ideas, and more engineers to be able to build things that would otherwise take far more time and money. I want the infrastructure that makes that possible. I think that's worth working toward, and I don't want us to put it on hold because the most visible uses of AI have made it easy to overlook everything else.